1.Technical Field
The present invention relates to an image processing technology for generating a high-resolution image from a low-resolution image.
2.Background Art
As a method for generating a high-resolution image from a low-resolution image, the super resolution technology has been proposed. For example, Non-Patent Literature 1, discloses a representative method of the super resolution technology. According to Non-Patent Literature 1, Equation 1 shown below is defined to represent relation between a plurality of low-resolution images and a high-resolution image, and the high-resolution image is generated by solving the Equation 1. To solve Equation 1, iterative processing using the gradient method or the like is performed, and the calculation is repeatedly performed until the error becomes constant.
                    I        =                                            1                              σ                2                                      ⁢                                          ∑                                  i                  =                  0                                                  N                                      l                    -                    1                                                              ⁢                                                [                                                                                                              b                          ⁡                                                      (                                                                                          x                                i                                                            ,                                                              y                                i                                                                                      )                                                                          T                                            ·                      h                                        -                                          f                      i                                                        ]                                2                                              +                      a            ⁢                                                                                                C                    T                                    ·                  h                                                            2                                                          (                  Equation          ⁢                                          ⁢          1                )            
In the above equation, h is a vector representation of a high-resolution image, σ is a noise deviation of an observed value, b(xi, yi) is a vector representation of a PSF (Point Spread Function) kernel corresponding to a position (xi, yi), C is a matrix representing prior information of the high-resolution image, and α is a constraint parameter representing the strength of the constraint. Also, Nl is the total number of pixels constituting a plurality of low-resolution images.
In conventional super resolution technologies, based on an assumption that a low-resolution image is a result of a high-resolution image having been degraded for some reason, the relation between a target high-resolution image and a low-resolution image is described in an equation, and the target high-resolution image is obtained by solving the equation. In obtaining an excellent high-resolution image, the equation as a degradation model plays a vital role, and how to create a degradation model of a low-resolution image based on an assumed degradation process is a key to the super resolution technology.
When the degradation model differs from the real degradation process of the low-resolution image, a noise is included in the obtained high-resolution image due to the difference between the degradation model and the real degradation process. Non-Patent Literature 2 discloses a super resolution technology that enhances the robustness by adding a member representing a system noise into the degradation model, other than the members representing noises occurring in the process of inputting the image such as the PSF and the like. According to the technology of Non-Patent Literature 2, a degradation model such as Equation 2 is defined, and a high-resolution image is generated by solving this equation.Yk=DkHkcamFkHkatmX+Vk k=l, . . . ,N  (Equation 2)
In Equation 2, X is an input image, Hatm is an atmospheric blur, Fk is a degradation caused by A/D conversion, Hcam is a camera blur, Dk is a degradation caused by downsampling, and Vk is a system noise. In Equation 2, the degradation including the system noise is modeled. Thus the equation provides robustness with regard to noises other than the noises of the regular image degradation.